Technologies For Adding Resiliency to Dynamic Distributed Systems

Università di Roma “La Sapienza” Dipartimento di Informatica e Sistemistica Middleware Laboratory MIDLAB

Technologies For Adding Resiliency to Dynamic Distributed Systems Roberto Baldoni

Sapienza Univ. Rome, Sapienza Innovazione

Cortina – 31/1/2009

Mid

dlew

are

Labo

rato

ry

MID

LAB

Initial Statement

Distributed systems are rapidly evolving, into new classes of applications such as VANET, Smart Environments, P2P, and distributed cloud services

Mastering this complexity is an industry need

Common denominator: Dynamicity of the system model e.g. Underlying communication structure self-defined at any

instant by entities

Continuous arrival and departure of participating entities

it may not be possible to assume anything about the universe of participants, their identities, capabilities, or reliability.

Mid

dlew

are

Labo

rato

ry

MID

LAB

What is a dynamic distributed systems?

Internet-scale Applications

Enterprise Data Centers

Scalable QoS-constrained applications

First Open Workshop Budapest 21-3-2007

Mid

dlew

are

Labo

rato

ry

MID

LAB

What is a dynamic distributed systems?

Internet-scale Applications •  unmanaged environment •  shortlife peers •  High churn

Enterprise Data centers •  managed environment •  longlife peers •  low churn

Scalable QoS-Constrained Application •  partially managed environment •  shortlife peers at network edges, longlife peers in the core

•  high churn only at network edges, low churn in the core

Very different system models

Mid

dlew

are

Labo

rato

ry

MID

LAB

Each application has requirements that impact the design of the overlay

• self-organize and partial view of the whole system • gracefully tolerate peer failures and churn • avoiding single point failures and bottlenecks

Ideal software substrate for dynamic distributed systems?

Mid

dlew

are

Labo

rato

ry

MID

LAB

Overlay Networks Substrate as superimposion of graphs

Structured overlay

Unstructured overlay

Hybrid overlay

ring

hypercube

ring

small-world

Pastry [RD01]

Chord [SMKKB03] random

Skip-list

GosSkip [GHHKFR06]

ring

random ring

BISE [TCS06]

tree

Mid

dlew

are

Labo

rato

ry

MID

LAB

Looking what’s going on

P2P technology has started in Internet based applications such as file sharing and IP telephony a-la-skype but there has not be wide diffusion of other applications due to the difficulty of handling the dynamicity of the internet based model

This technology has much more potential if looking at domain specific enviroments where some of the dimension of dynamicity can be managed by some provider.

Mid

dlew

are

Labo

rato

ry

MID

LAB

Smart Houses (The SM4ALL Project):

Future scenarios for superimposed Overlay network

Mid

dlew

are

Labo

rato

ry

MID

LAB

Financial Infrastructures (the CoMiFin Project):

Future scenarios for superimposed Overlay network

Mid

dlew

are

Labo

rato

ry

MID

LAB

SOFIA (Artemis 30Me Project in FP7 led by NOKIA). Smart Objects for Intelligent applications

eDiana (Artemis 15Me Project in FP7 led by Acciona, Spain) on cooperative building for energy saving.

Future scenarios for superimposed Overlay network

Mid

dlew

are

Labo

rato

ry

MID

LAB

Conclusion

Understanding the fundamentals of how to master this �dynamic dimension is of primary importance to design of robust, dependable, and predictable distributed systems.

From the theory side (PODC/DISC) we have basically no common background on that (Theoretical Aspects of Dynamic Distributed Systems Workshop will be held in September in Spain)

Problem coming from industry, technology leading, lack of shared theoretical background

There are all the ingredients for having fun in the next years!!